Loading your bets…
"Over up to 19.5" means the over is worth betting at 19.5 or any lower line at that price. "Worth betting" means at least $3 expected profit per $100.
Weather: forecasts for outdoor games are entered in each player's calculator. Wind of 15+ mph and temperatures below 40°F reduce pass attempts, which also lowers receptions and passing yards. Domes are not affected.
Each market was tested on a season it wasn't fit on. "Model said" is the predicted chance of the over; "Actually hit" is how often it happened. Closer is better.
A factor stays in only if it improved projections when fit on 2024 and tested on 2025, and the reverse. Kept: spread, total, team and opponent tendencies, opponent run defense, three-and-out rate, no-huddle rate, and run efficiency for volume; snap counts and injury redistribution for RB carries; cold and strong wind for pass attempts; opponent completion rate allowed and implied team total for passing yards. Left out: most other defensive stats, snap counts for receivers, redistributing targets when a receiver is out, and defensive stats for rushing and receiving yards.
Projected winners and win chances for this week's games. Picks follow the betting market with the bookmaker's cut removed, which picked the winner 68% of the time in 2023–2025. Tap a team to check bet365's price.
I built team power ratings from play-by-play efficiency and tested them on the 2023–2025 seasons against real closing moneylines. The market picked the winner of the time; the ratings picked . Betting every game where the ratings disagreed with the market lost per bet, and mixing the two did no better. So the ratings appear only as a second opinion.
Where moneylines can pay is price. When bet365 offers better odds than the fair price, the bet has positive expected value. The calculator shows a bet only when it's worth at least $3 per $100. Check a second book too, because the fair price here comes from one consensus source.
A big disagreement between the ratings and the market usually means news, most often a quarterback change the ratings don't know about. Games with a new listed starter are flagged.
How often favorites actually won at each fair chance, 2023–2025 ( games):
Loading your bets…
Checking whether Claude can read screenshots in this view…
"Over up to 24.5" means the over is worth betting at 24.5 or any lower line at that price. "Worth betting" means at least $3 expected profit per $100.
Each projection is projected minutes × the player's per-minute production. Minutes come from his recent minutes, his last game, and how crowded his team's rotation is. When you mark a teammate out, his minutes spread to everyone else and scorers get a usage bump, both amounts learned from the last two seasons.
The model can't see injury reports or rest decisions. Check the report before betting and tap Mark player out in the calculator. Outs you mark apply to that game day only.
Fit on one season and tested on the other (2024–25 and 2025–26), in both directions. "Model said" is the predicted chance of the over; "Actually hit" is how often it happened.
Kept after testing: recent minutes, last game's minutes, rotation crowding, teammates out (points), and opponent assists allowed (assists). Tested and left out because they didn't help in both seasons: pace, opponent points, rebounds and threes allowed, home court, back-to-backs and blowout risk for player stats, and starter status.
Projected winners from team ratings (points scored and allowed per 100 possessions, recent games weighted most) plus home court and back-to-backs.
In 2024–25 and 2025–26, the ratings picked the winner of the time, and their win chances matched how often teams actually won. I couldn't compare them with betting lines, because free NBA data doesn't include historical odds. In the NFL test, the betting market beat team ratings, so treat these as a guide, not a betting signal.
Early in the season the ratings lean on last season. Teams that changed a lot over the summer will be off until about 15–20 games in.
Loading your bets…
Checking whether Claude can read screenshots in this view…
"Over up to 24.5" means the over is worth betting at 24.5 or any lower line at that price. "Worth betting" means at least $3 expected profit per $100.
Each projection is projected minutes × the player's per-minute production. Minutes come from her recent minutes, her last game, and how crowded her team's rotation is. In the playoffs, starters' minutes go up and the bench's go down, an adjustment learned from the last two postseasons. When you mark a teammate out, her minutes spread to everyone else.
The model can't see injury reports. Check the report before betting and tap Mark player out in the calculator. Outs you mark apply to that game day only.
Fit on one season and tested on the other (2025 and 2026), in both directions. "Model said" is the predicted chance of the over; "Actually hit" is how often it happened.
Kept after testing: recent minutes, last game's minutes, rotation crowding, and the playoff minutes adjustment. Tested and left out because they didn't help in both seasons: pace, opponent points, rebounds, assists and threes allowed, a scoring bump when teammates are out, home court, back-to-backs, blowout risk and starter status.
Projected winners from team ratings (points scored and allowed per 100 possessions, recent games weighted most) plus home court.
In the 2025 and 2026 seasons, the ratings picked the winner of the time, and their win chances matched how often teams actually won. I couldn't compare them with betting lines, because free NBA data doesn't include historical odds. In the NFL test, the betting market beat team ratings, so treat these as a guide, not a betting signal.
Playoff series are short and teams adjust game to game, so treat close calls as coin flips.
Loading your bets…
Checking whether Claude can read screenshots in this view…